Speech Emotion Recognition Based on EMD in Noisy Environments

Abstract

In the speech emotion recognition process, How to obtain effective characteristic parameters from the emotional data including the noise is one of the significant and difficult problem. This paper first removes the gauss white noise with the adaptive filter. Then the Mel Frequency Cepstrum Coefficients (MFCC) based on Empirical Mode Decomposition (EMD) is extracted and with its difference parameter to improve. At last we present an effective method for speech emotion recognition based on Fuzzy Least Squares Support Vector Machines (FLSSVM) so as to realize the speech recognition of four main emotions, i.e, anger, happy, surprise and natural. The experiment results show that this method has the better anti-noise effect when compared with traditional Support Vector Machines (SVM).